Part 3 – Binary outcome,

difference in proportions
between groups
 In this Section
• Focus on binary outcome variable
• Find differences between the groups
(is the proportion of the disease higher in
one group than in another??)
– Relative risk
– Odds ratio
– χ2-test
– Logistic regression
 Organize categorical data
2 x 2 table
Disease Disease
Total
Yes/+ No/
Exposure
a b (a+b)
Yes/+
Exposure
c d (c+d)
No/
(a+b+c+d)
Total (a+c) (b+d)
=N
 Example

Diabetes Diabetes
Totalt
+ 
Smoke + a=14 b=976 990
Smoke  c=19 d=777 796
Totalt 33 1753 1786
Risk
• The Risk is the
probability of an event

• Risk of the disease in each group:

– Probability (exposed) = a/(a+b)
– Probability (not exposed) = c/(c+d)

• Relative risk of disease,

exposed vs. not exposed:
– RR = a/(a+b) / c/(c+d) = a(c+d) / c(a+b)
Risk Diabetes
+
Diabetes

Totalt

Smoke + a=14 b=976 990

• The Risk is the
probability of an event Smoke  c=19 d=777 796

Totalt 33 1753 1786

• Risk of the disease in each group:

– Probability (exposed) = a/(a+b) = 14/990 = 0.0141
– Probability (not exposed) = c/(c+d) = 19/796 = 0.0239

• Relative risk of disease,

exposed vs. not exposed:
– RR = a/(a+b) / c/(c+d) = a(c+d) / c(a+b) =0.0141/0.0239 = 0.592
Odds
• Odds is the likelihood
of an event divided by
the likelihood of a non-event

• Odds of the disease in each group:

– Odds (exposed)=a/(a+b) / b(a+b) = a/b
– Odds (not exposed)=c/(c+d) / d(c+d) = c/d
• Odds of the disease,
exposed vs. not exposed:
– OR = (a/b)/(c/d) = a*d/b*c
Odds Diabetes
+
Diabetes

Totalt

Smoke + a=14 b=976 990

• Odds is the likelihood Smoke  c=19 d=777 796

of an event divided by Totalt 33 1753 1786
the likelihood of a non-event

• Odds of the disease in each group:

– Odds (exposed)=a/(a+b) / b(a+b) = a/b = 14/976 = 0.0143
– Odds (not exposed)=c/(c+d) / d(c+d) = c/d = 19/777 = 0.0245
• Odds of the disease,
exposed vs. not exposed:
– OR = (a/b)/(c/d) = a*d/b*c = 0.0143/0.0245 = 0.587
 Compare proportions in groups

Calculating RR and OR in SPSS can be quite challenging

• Go to «Analyze> Descriptive statistics> Crosstabs»
Compare proportions in groups
The outcome should go in the column
and the exposure in the row,
• The outcome we are interested in
must be coded with the lowest value
• The exposure we are interested in
must be coded with the lowest value

• It can be necessary to recode the variables

 Exercise 3a
• Find a possible connection between
being a smoker (the variable «smoking»)
and myocardial infarction (the variable «mi»)
by calculating RR and OR

Hint: Re-code smoking first

 Exercise 3a: Solution
• Re-code «smoking» to
«smokingNew» with
two categories.
 Exercise 3a: Solution
• Re-code «smoking» to
«smokingNew» with
two categories.
– Remember: The category
we are interested in (those
who have smoked) should
have the lowest value.
 Exercise 3a: Solution
• Re-code «smoking» to
«smokingNew» with
two categories.
– Remember: The category
we are interested in (those
who have smoked) should
have the lowest value
– Add «Value Lables».
• Also MI should be re-coded
«mi»  «miNew»
Add Value Labels (Yes / No)
 Exercise 3a: Solution
• Go to «Analyze >Descriptive
Statistics >Crosstabs».
• «smokingNew» as rows
«miNew» as columns
 Exercise 3a: Solution
• Go to «Analyze >Descriptive
Statistics >Crosstabs».
• «smokingNew» as rows
«miNew» as columns
• Click «Statistics», and select
«Risk». Then «Continue»
 Exercise 3a: Solution
• Go to «Analyze >Descriptive
Statistics >Crosstabs».
• «smokingNew» as rows
«miNew» as columns
• Click «Statistics», and select
«Risk». Then «Continue»
• Click «Cells», select
«Observed» in «Counts»,
«Row» in «Percentages».
Then «Continue».
 Exercise 3a: Solution
• Go to «Analyze >Descriptive
Statistics >Crosstabs».
• «smokingNew» as rows
«miNew» as columns
• Click «Statistics», and select
«Risk». Then «Continue»
• Click «Cells», select
«Observed» in «Counts»,
«Row» in «Percentages».
Then «Continue».
• Click «OK» in the original
dialog box.
 Exercise 3a: Solution
Interpreting output:
• The odds are 2,3 times
higher (95% CI: 1,6-3,4)
for smokers, compared
to non-smokers
 Exercise 3a: Solution
Interpreting output:
• The odds are 2,3 times
higher (95% CI: 1,6-3,4)
for smokers, compared
to non-smokers

• For comparison,
the relative risk
(317/1491) / (31/295) =
2,02
 RR and OR
• RR og OR are useful tools to assess the effect
of a binary exposure on a binary outcome
– RR is easier to understand
– OR sometimes is more useful
• RR makes no sense in case-control studies, OR does
• OR is used in logistic regression i logistisk regresjon
– OR is always more extreme than RR
• OR>RR>1
• OR<RR<1
• OR=RR=1
• If the outcome is rare, then OR≈RR
 χ -test
2

If we are only interested in finding if there is a

correlation between the row variable (exposure) and
the column variable (outcome), without wanting to
quantify the difference between the exposure groups,
we can use a χ2-test.
 χ -test
2

If we are only interested in finding if there is a

correlation between the row variable (exposure) and
the column variable (outcome), without wanting to
quantify the difference between the exposure groups,
we can use a χ2-test.
• Can have more than 2 categories in both variables.
 χ -test
2

• Base Idea: comparing what we observe with what we would

expect if we know that the variables are uncorrelated.

– The expected numbers are obtained from

the row and column totals in the table.
– Compares the expected numbers with
the observed numbers for each cell.

• Variable coding is not influent for SPSS.

 χ -test
2

We want to find out if there is a

connection between BMI
categories and diabetes.
- «bmicat» (exposure)
- «diabetes» (outcomes)
 χ -test
2

We want to find out if there is a

connection between BMI
categories and diabetes.
• Go to «Analyze >Descriptive
Statistics >Crosstabs»
Add «bmicat» in «Rows»,
and «diabetes» in «Columns».
 χ -test
2

We want to find out if there is a

connection between BMI
categories and diabetes.
• Go to «Analyze >Descriptive
Statistics >Crosstabs».
• Click «Statistics», and select
«Chi-square», then «Continue»
 χ -test
2

We want to find out if there is a

connection between BMI
categories and diabetes.
• Go to «Analyze >Descriptive
Statistics >Crosstabs».
• Click «Statistics», and select
«Chi-square», then «Continue»
• Click «Cells», and select
«Observed» and «Expected»
under «Counts» and «Row»
under «Percentages».
Then «Continue»...
 χ -test
2

We want to find out if there is a

connection between BMI
categories and diabetes.
• Go to «Analyze >Descriptive
Statistics >Crosstabs».
• Click «Statistics», and select
«Chi-square», then «Continue»
• Click «Cells», and select
«Observed» and «Expected»
under «Counts» and «Row»
under «Percentages».
Then «Continue» and «OK».
 χ -test
2

Interpreting output:
• First table:
– Observed counts.
 χ -test
2

Interpreting output:
• First table:
– Observed counts.
– Expected counts
(if there was no difference
in the BMI categories).
 χ -test
2

Interpreting output:
• First table:
– Observed counts.
– Expected counts
(if there was no difference
in the BMI categories).

– The proportion
with diabetes in
each category.
 χ -test
2

Interpreting output:
• First table:
– Observed counts.
– Expected counts
(if there was no difference
in the BMI categories).

– The proportion
with diabetes in
each category.
• The second table gives
the result of the χ2-test.
 χ -test
2

Interpreting output:
• First table:
– Observed counts.
– Expected counts
(if there was no difference
in the BMI categories).

– The proportion
with diabetes in
each category.
• The second table gives
the result of the χ2-test.
- BMI does not have an
effect on Diabetes
 χ -test
2

To use the χ2-tes for a 2x2-table:

• Total number of units in the table/study have to be above 40
OR
• Total number of units must be between 20 and 40,
and all expected values are greater than 5

To use the χ2-tes for a table bigger than 2x2:

• Less than 20% of the expected values are lower than 5
AND
• No expected values are less than 1
χ -test
2
 Logistic regression
• Similar to linear regression, but with binary outcome.
 Logistic regression
• Similar to linear regression, but with binary outcome.
– We can adjust for categorical variables (2+ categories).
 Logistic regression
• Similar to linear regression, but with binary outcome.
– We can adjust for categorical variables (2+ categories).
– We can adjust for continuous variables.
 Logistic regression
• Similar to linear regression, but with binary outcome.
– We can adjust for categorical variables (2+ categories).
– We can adjust for continuous variables.
But:
• We calculate the logarithm of an odds ratio, rather
than a continuous variable.
 Logistic regression
• Similar to linear regression, but with binary outcome.
– We can adjust for categorical variables (2+ categories).
– We can adjust for continuous variables.
But:
• We calculate the logarithm of an odds ratio, rather
than a continuous variable.

Instead of
 Logistic regression
• Similar to linear regression, but with binary outcome.
– We can adjust for categorical variables (2+ categories).
– We can adjust for continuous variables.
But:
• We calculate the logarithm of an odds ratio, rather
than a continuous variable.

Instead of

p/(1-p) is the odds ratio of the outcome

 Logistic regression

• If is a binary variable (for example smoking):

– if smoker, og if not smoker.
– The odds of smokers is .
– The odds of non smokers is
– OR is: .
• The exponential of the coefficient, , is
interpretable as an OR!
 Logistic regression

• If is a binary variable (for example smoking):

– if smoker, og if not smoker.
– The odds of smokers is .
– The odds of non smokers is
– OR is: .
• The exponential of the coefficient, , is
interpretable as an OR!
 Logistic regression

• If is a binary variable (for example smoking):

– if smoker, og if not smoker.
– The odds of smokers is .
– The odds of non smokers is
– OR is: .
• The exponential of the coefficient, , is
interpretable as an OR!
 Logistic regression

• If is a binary variable (for example smoking):

– if smoker, og if not smoker.
– The odds of smokers is .
– The odds of non smokers is
– OR is: .
• The exponential of the coefficient, , is
interpretable as an OR!
 Logistic regression

• If is a binary variable (for example smoking):

– if smoker, og if not smoker.
– The odds of smokers is .
– The odds of non smokers is
– OR is: .
• The exponential of the coefficient, , is
interpretable as an OR!
 Logistic regression
• SPSS:
– Interprets the highest value of the (sick / not sick)
as the one we are interested in.
 Logistic regression
• SPSS:
– Interprets the highest value of the (sick / not sick)
as the one we are interested in.
– For the independent variable (exposure), we select
the reference category (baseline) in the menu.
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
• Original diabetes variable
(1=diabetes, and 0= no diabetes)
as dependent variable (outcome variable).
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
• Original diabetes variable
(1=diabetes, and 0= no diabetes)
as dependent variable (outcome variable).
• Adds «overvekt» for «Covariates».
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
• Original diabetes variable
(1=diabetes, and 0= no diabetes)
as dependent variable (outcome variable).
• Adds «overvekt» for «Covariates».
Then, click «Categorial».
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
• Original diabetes variable
(1=diabetes, and 0= no diabetes)
as dependent variable (outcome variable).
• Adds «overvekt» for «Covariates».
Then, click «Categorial».
– Move «overvekt» in «Categorical Covariates».
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
• Original diabetes variable
(1=diabetes, and 0= no diabetes)
as dependent variable (outcome variable).
• Adds «overvekt» for «Covariates».
Then, click «Categorial».
– Move «overvekt» in «Categorical Covariates».
– Select «Reference Category: Last» .
– Then, «Continue».
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
• Original diabetes variable
(1=diabetes, and 0= no diabetes)
as dependent variable (outcome variable).
• Adds «overvekt» for «Covariates».
Then, click «Categorial».
– Move «overvekt» in «Categorical Covariates».
– Select «Reference Category: Last» .
– Then, «Continue».

NB! If you want to change the reference

category, remember to press «Change»
before pressing «continue»!
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
• Original diabetes variable
(1=diabetes, and 0= no diabetes)
as dependent variable (outcome variable).
• Adds «overvekt» for «Covariates».
Then, click «Categorial».
– Move «overvekt» in «Categorical Covariates».
– Select «Reference Category: Last» .
– Then, «Continue».
• Click «Options», then select:
– «CI for exp(B): 95%»
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
• Original diabetes variable
(1=diabetes, and 0= no diabetes)
as dependent variable (outcome variable).
• Adds «overvekt» for «Covariates».
Then, click «Categorial».
– Move «overvekt» in «Categorical Covariates».
– Select «Reference Category: Last» .
– Then, «Continue».
• Click «Options», then select:
– «CI for exp(B): 95%»
– «Include constant in model»
– Then, «Continue».
 Logistic regression - example
Estimate logistic model for «diabetes»,
with «overvekt» as independent variable.
(overvekt =1, not overvekt =2)
• Click «Analyze->Regression->Binary logistic».
• Original diabetes variable
(1=diabetes, and 0= no diabetes)
as dependent variable (outcome variable).
• Adds «overvekt» for «Covariates».
Then, click «Categorial».
– Move «overvekt» in «Categorical Covariates».
– Select «Reference Category: Last» .
– Then, «Continue».
• Click «Options», then select:
– «CI for exp(B): 95%»
– «Include constant in model»
– Then, «Continue».
• Click «OK» in the original box.
 Logistic regression - example
Interpreting output:
• SPSS provides a lot of info
– Start by checking that the
numbers are correct
 Logistic regression - example
Interpreting output:
• SPSS provides a lot of info
– Start by checking that the
numbers are correct

– The estimation of the model is in

the last table!
Logistic regression - example
• We see from the table that exp(B)
is 1,8 (95% KI: 0,8 – 4,0).
Logistic regression - example
• We see from the table that
exp(B) is 1,8 (95% KI: 0,8 – 4,0).
– Odd of having diabetes is
1,8 times bigger for overweight
compared to non-overweight.
Logistic regression - example
• We see from the table that
exp(B) is 1,8 (95% KI: 0,8 – 4,0).
– Odd of having diabetes is
1,8 times bigger for overweight
compared to non-overweight.
– NB! Not significant.
 Logistic regression
We can have an independent variable
(exposure) with multiple categories.
 Logistic regression
We can have an independent variable
(exposure) with multiple categories.
• SPSS designs the «dummy-variabler» for us.
 Logistic regression
We can have an independent variable
(exposure) with multiple categories.
• SPSS designs the «dummy-variabler» for us.
– We choose the baseline (reference category).
 Logistic regression
We can have an independent variable
(exposure) with multiple categories.
• SPSS designs the «dummy-variabler» for us.
– We choose the baseline (reference category).
– Calculate a coefficient for each category.
 Logistic regression
We can have an independent variable
(exposure) with multiple categories.
• SPSS designs the «dummy-variabler» for us.
– We choose the baseline (reference category).
– Calculate a coefficient for each category.
– is the OR compared to the baseline.
 Logistic regression
We can have an independent variable
(exposure) with multiple categories.
• SPSS designs the «dummy-variabler» for us.
– We choose the baseline (reference category).
– Calculate a coefficient for each category.
– is the OR compared to the baseline.
• For a variable with 3 categories:
 
 Logistic regression
We can have an independent variable
(exposure) with multiple categories.
• SPSS designs the «dummy-variabler» for us.
– We choose the baseline (reference category).
– Calculate a coefficient for each category.
– is the OR compared to the baseline.
• For a variable with 3 categories:
 

Is the OR of the 1st category compared to cat-0

 Logistic regression
We can have an independent variable
(exposure) with multiple categories.
• SPSS designs the «dummy-variabler» for us.
– We choose the baseline (reference category).
– Calculate a coefficient for each category.
– is the OR compared to the baseline.
• For a variable with 3 categories:
 

Is the OR of the 2nd category compared to cat-0

 Exercise 4b
Build a logistic regression model for
myocardial infarction («mi»), with
smoking as an independent variable.
Use the variable «smoking»,
which has 5 categories.
Select «Never smoked» (value 0)
as reference category.
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
• Add «smoking» in «Covariates».
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
• Add «smoking» in «Covariates».
• Then, click «Categorial».
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
• Add «smoking» in «Covariates».
• Then, click «Categorial».
- Put «smoking» in «Categorical Covariates»
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
• Add «smoking» in «Covariates».
• Then, click «Categorial».
- Put «smoking» in «Categorical Covariates»
- Remember: 0 = ‘Never smoked’
- Select «Reference Category: First»
and click «Change».
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
• Add «smoking» in «Covariates».
• Then, click «Categorial».
- Put «smoking» in «Categorical Covariates»
- Remember: 0 = ‘Never smoked’
- Select «Reference Category: First»
and click «Change».
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
• Add «smoking» in «Covariates».
• Then, click «Categorial».
- Put «smoking» in «Categorical Covariates»
- Remember: 0 = ‘Never smoked’
- Select «Reference Category: First»
and click «Change».
• Click «Options».
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
• Add «smoking» in «Covariates».
• Then, click «Categorial».
- Put «smoking» in «Categorical Covariates»
- Remember: 0 = ‘Never smoked’
- Select «Reference Category: First»
and click «Change».
• Click «Options».
– Check for «CI for exp(B): 95%».
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
• Add «smoking» in «Covariates».
• Then, click «Categorial».
- Put «smoking» in «Categorical Covariates»
- Remember: 0 = ‘Never smoked’
- Select «Reference Category: First»
and click «Change».
• Click «Options».
– Check for «CI for exp(B): 95%».
– «Include constant in model» should be selected.
Click «Continue».
 Exercise 4b: Solution
Logistic regression model for «mi», with
«smoking» as independent variable.
• «Analyze> Regression> Binary logistic»
• The original variable as «Dependent».
• Add «smoking» in «Covariates».
• Then, click «Categorial».
- Put «smoking» in «Categorical Covariates»
- Remember: 0 = ‘Never smoked’
- Select «Reference Category: First»
and click «Change».
• Click «Options».
– Check for «CI for exp(B): 95%».
– «Include constant in model» should be
selected. Click «Continue».
• Click «OK» in the original dialog box.
 Exercise 4b: Solution
Interpreting output:
 Exercise 4b: Solution
Interpreting output:
• Make sure you have chosen a reasonable
reference group!
 Exercise 4b: Solution
Interpreting output:
• OR with CI for each category
compared to categoty 0.
 Logistic regression
• We can also have a continuous outcome
 Logistic regression
• We can also have a continuous outcome

• the interpretation of will be:

 Logistic regression
• We can also have a continuous outcome

• the interpretation of will be:

- OR associated to the increase of one unit
of the independent variable.
 Logistic regression
• We can also have a continuous outcome

• the interpretation of will be:

- OR associated to the increase of one unit
of the independent variable.
- If 𝑥1 is weight in kg, exp(β1) is the OR that
compares an individual with a specific weight,
with an individual weighing one kg less.
NOTE: OR is the same when comparing someone who
weighs 2kg vs. 1kg, like 102kg vs. 101kg.
 Exercise 4c
Estimate a logistic regression model for
myocardial infarction («mi»), using the
continuous BMI variable («bmi») as independent
variable (exposure).
 Exercise 4c: Solution
Logistic regression model for «mi»,
with «bmi» as independent variable.
• «Analyze>Regression>
Binary logistic»

–.
–.
 Exercise 4c: Solution
Logistic regression model for «mi»,
with «bmi» as independent variable.
• «Analyze>Regression>
Binary logistic»
• The original variable as
«Dependent».

–.
–.
 Exercise 4c: Solution
Logistic regression model for «mi»,
with «bmi» as independent variable.
• «Analyze>Regression>
Binary logistic»
• The original variable as
«Dependent».
• Add «bmi» in «Covariates».

–.
–.
 Exercise 4c: Solution
Logistic regression model for «mi»,
with «bmi» as independent variable.
• «Analyze>Regression>
Binary logistic»
• The original variable as
«Dependent».
• Add «bmi» in «Covariates».
• Click «Options».
–

–.
–.
 Exercise 4c: Solution
Logistic regression model for «mi»,
with «bmi» as independent variable.
• «Analyze>Regression>
Binary logistic»
• The original variable as
«Dependent».
• Add «bmi» in «Covariates».
• Click «Options».
– Check for «CI for exp(B): 95%».

–.
–.
 Exercise 4c: Solution
Logistic regression model for «mi»,
with «bmi» as independent variable.
• «Analyze>Regression>
Binary logistic»
• The original variable as
«Dependent».
• Add «bmi» in «Covariates».
• Click «Options».
– Check for «CI for exp(B): 95%».
– «Include constant in model» should
be selected.
–.
 Exercise 4c: Solution
Logistic regression model for «mi»,
with «bmi» as independent variable.
• «Analyze>Regression>
Binary logistic»
• The original variable as
«Dependent».
• Add «bmi» in «Covariates».
• Click «Options».
– Check for «CI for exp(B): 95%».
– «Include constant in model» should
be selected.
– Click «Continue».
 Exercise 4c: Solution
Logistic regression model for «mi»,
with «bmi» as independent variable.
• «Analyze>Regression>
Binary logistic»
• The original variable as
«Dependent».
• Add «bmi» in «Covariates».
• Click «Options».
– Check for «CI for exp(B): 95%».
– «Include constant in model» should be
selected.
– Click «Continue».
• Click «OK» in the original box.
 Exercise 4c: Solution
Interpreting output:
• OR of 1.05 (95% CI) associated with one unit
increase in BMI.
 Exercise 4c: Solution
Interpreting output:
• OR of 1.05 (95% CI) associated with one unit
increase in BMI.
• OR associated with an increase in BMI of 10:
exp(10*0.051) = Exp(0.51) = 1.67
 Summary
• Compare differences of proportions of two groups.
RR or OR.
Easier way: use logistic regression in SPSS, even if
you have only one independent variable with two levels
 Compare the difference in proportions of more than two
groups without quantifying the difference.
χ2-test.
 Compare the difference in proportions of more than two
groups, with the wish to quantify the difference.
Logistic regression.
• Adjust for other variables.
Logistic regression.